Stream model reserve to Scaleway without local weight staging

Assistant: codex
Assistant-Model: gpt-6-astra
Assistant-Session: 01a09cbd-43c1-79f3-809e-1ee97b40b64d
This commit is contained in:
tegwick 2026-09-14 00:17:15 +02:00
parent ad88d16a52
commit 8a65297d56
16 changed files with 805 additions and 145 deletions

View file

@ -217,15 +217,20 @@ Operator / railiance-platform:
```task
id: FI-WP-0004-T09
status: todo
status: in_progress
priority: medium
state_hub_task_id: "304816de-5065-5928-b9bd-da1b5019c3d4"
```
After T08: pull `deepseek-ai/DeepSeek-V4-Flash-0731` via
`scripts/collect_model.py` into staging, upload as `strategic/` Glacier
(or Standard then lifecycle). Verify MANIFEST vs objects. Update catalog
`status: collected` / `verified` and RESERVE-STATUS.
**2026-09-14:** The old WSL download is stopped/inactive. Collection now runs
as `fi-reserve-collect.service` on Railiance using diskless streaming (64 MiB
parts, 512 MiB memory cap, zero swap). Source pinned to
`7872f01b1d1fe23eabc4c98b48bffcef5a386062`; 67 files / 166,898,547,054 bytes.
Live multipart tests passed for STANDARD, ONEZONE_IA and GLACIER. Provider
ignored wrong Content-MD5, so explicit part/composite ETag validation is used.
See `docs/streaming-reserve.md`. Transfer completion and final manifest remain
pending. Update catalog `collected` only after the manifest is complete;
`verified` requires restore/readback.
**Done when:** catalog `storage_path` is the S3 prefix and checksums match.